Image completion refers to the problem of recovering the missing, corrupted or obscured entries in image data. In this paper, we consider the problem in the remote sensing domain, where regions of an image are missing due to difficulties such as cloud cover, sensor failures or partial sensor coverage. Where previous work in this field generally falls into the category of low-rank completion methods, we propose a novel graph-based diffusion approach to the problem. The method, referred to as GraphProp, propagates observed entries around a graph-based representation of the image region in order to recover the missing entries. The graph-based diffusion approach to completion is to the best of our knowledge a novel method for remote sensing image completion. Using real-world multispectral image data acquired from the Landsat 7 platform, we validate our approach using experiments which synthetically obscure image sections. In these tests, we benchmark against alternative image completion approaches and demonstrate the superior reconstruction performance of our method versus the state of the art. Code which implements the method has been made publicly available at https://github.com/iainrolland/GraphProp.

Classic graph neural networks show some limitations in information extraction performance when applied to multimodal datasets. This is primarily due to such datasets having high volume, variety, and variability. In this paper, we propose structuring graph neural networks on a new graph representation based on fluid dynamics diffusion that allows us to incorporate the reliability of the features used to characterise each sample within the graph structure itself. This approach aims to address some of the major limitations of the classic graph-based learning structures, so to improve accuracy and robustness of the estimates. We show how this approach can help to strongly improve the quality of the analysis of classic graph neural networks. Experimental results are reported to support this point.

Remote sensing can provide vital information in the aftermath of a disaster, and can be used by search and rescue teams to strategically deploy resources. However, there is little work on methods which address their operational requirement. This paper presents two methods: the first aims to deploy fast, robust information that can evolve its output as information becomes available; the second focuses on providing an estimation of the degree of confidence for the outcomes of a classification performed by a graph convolutional network. These methods are tested by using data collected following the Beirut blast in 2020.

High precision monitoring of infrastructure using artificial reflectors is possible with freely available Sentinel-1 data, but large reflectors are needed. We find that a triangular trihedral corner reflector should typically have at least 1 m inner leg length. As such large reflectors are often not feasible for use in urban areas for essential infrastructure monitoring, we designed a multiple corner-reflector array to replace a single corner reflector with an inner leg length of 1 m. In this case, we use four reflectors where each of them is a truncated triangular trihedral with an inner leg length of 0.33 m. We measured InSAR amplitude, phase and coherence of this reflector array with various configurations of alignments of the array. We find that as long as great care is taken in the relative positioning of the four corner reflectors, so that they constructively interfere, each horizontal or vertical configuration provides the expected amplitude, coherence and phase stability. Applications of multiple small corner reflectors in urban areas range from essential infrastructure monitoring (e.g bridges, overpasses, tunnel constructions), through assessment of structural health of buildings, to monitoring highway and railway embankments. We show that the multiple corner array works when placed in a single InSAR resolution cell, but depending on the application, the number and projection of corner reflectors can be varied, as long as sufficient signal-to-clutter ratio is achieved in the area of interest.

The recent advancements in machine learning techniques have opened the door for automatic large scale monitoring of the surface of the earth. For instance, they could be used in order to evaluate and assess civil infrastructures at scale, which is costly due to the fact that typically the existing methods rely on in-situ evaluation. Over the last decade Deep Learning technologies have risen as the state of the art methods for many different machine learning problems due to the fact that they can learn complex features and model complex non-linear behaviours. In this paper we will explore the possibility of using Deep Learning technologies over remote sensing data with the aim of structure health monitoring at scale. We will compare the performance of new Deep Learning technologies with regards to other traditional machine learning methods. For this purpose, we will use InSAR (Interferometry Synthetic Aperture Radar) data which allow us to measure cumulative surface displacement in the line of sight of the sensor with millimetric accuracy. We will analyse multi temporal InSAR data in order to model ground subsidence. In this paper we will discuss how deep learning technologies can learn to detect terrain subsidence over multi-temporal InSAR data automatically, providing much better results than traditional methods.

Bridge owners are faced with the challenge of maintaining an ageing and deteriorating asset portfolio. There is an increasing amount of InSAR data becoming available for built environment monitoring, with the opportunity to leverage free ESA Sentinel-1 data for regular monitoring. By understanding the potential to augment existing modelling tools used in common bridge engineering practice, InSAR can provide valuable additional insights and spot potential problems. In this paper we study Hammersmith Flyover, a concrete bridge in London, United Kingdom. We combine geometric models, structural models and InSAR data within a GIS environment and demonstrate how these systems can be used to regularly monitor specific behaviors including thermal expansion.

This article examines advances in interferometric synthetic aperture radar (InSAR) satellite measurement technologies to understand their relevance, utilization, and limitations for bridge monitoring. Waterloo Bridge is presented as a case study to explore how InSAR data sets can be combined with traditional measurement techniques including sensors installed on the bridge and automated total stations. A novel approach to InSAR bridge monitoring was adopted by the installation of physical reflectors at key points of structural interest on the bridge, in order to supplement the bridge's own reflection characteristics and ensure that the InSAR measurements could be directly compared and combined with in situ measurements. The interpretation and integration of InSAR data sets with civil infrastructure data are more than a trivial task, and a discussion of uncertainty of measurement data is presented. Finally, a strategy for combining and interpreting varied data from multiple sources to provide useful insights into each of these methods is presented, outlining the practical applications of this data analysis to support wider monitoring strategies.

Widespread deterioration of ageing infrastructure and the recent collapses of bridge structures highlight the vital importance of structural health monitoring. Satellite Interferometric Synthetic Aperture Radar (InSAR) provides remote measurements at millimetre scale over large areas, but the interpretation of such data into civil engineering contexts remains more than a trivial task. This work presents the monitoring of Waterloo Bridge to investigate the potential for InSAR to be used to support bridge management activities. Corner reflectors and target prisms for automated total station measurements have been installed at key points of interest as part of a wider structural monitoring system, and displacement data from traditional and satellite monitoring systems are compared and studied.

Failure of bridges due to scour is of great concern to bridge asset owners, and is currently very difficult to predict and monitor regularly using conventional assessment methods. This paper presents evidence of how InSAR techniques can be used to monitor bridges at risk of scour, using Tadcaster Bridge, England, as a case study. Tadcaster Bridge suffered a partial collapse due to river scour on the evening of December 29th, 2015 following a period of severe rainfall and flooding. SAR scenes over the bridge from the two-year period prior to the collapse are analysed using SBAS interferometry methods, highlighting a distinct movement in the region of the bridge where the collapse occurred prior to the actual event. This precursor to failure observed in the data suggests the possible use of InSAR in structural health monitoring of bridges at risk of scour, as a means of an early warning system.